AI share of voice (AI SOV) is the percentage of all brand mentions in AI-generated answers that belong to your brand rather than a competitor. It answers the question every marketing lead now gets asked: when buyers ask ChatGPT, Gemini or Perplexity for a shortlist, how often are we on it — and how big is our slice?
Most guides stop at the definition. This one gives you the exact formula (plus two weighted variants that can swing your score by double digits on identical data), a fully worked calculation you can rebuild in a spreadsheet, and benchmark distributions by category drawn from 2.4 million AI answers MaxAEO sampled between January 1 and May 31, 2026 — so you can judge whether your number is genuinely good or just sounds good.
What Is AI Share of Voice?
AI share of voice is the share of brand mentions your company captures in AI answers, measured against a fixed set of competitors, across a defined set of prompts and platforms. If AI answers name brands 1,000 times across your tracked prompts and 200 of those mentions are yours, your AI SOV is 20%.
The metric matters because AI answers behave like zero-sum shortlists. A ChatGPT response to "best CRM for a 50-person sales team" typically names three to seven vendors; every mention a competitor wins is consideration you lose. And the audience is no longer niche: 34% of US adults have used ChatGPT, roughly double the 2023 share, per Pew Research — and Gartner predicted traditional search volume would fall 25% by 2026 as that behavior shifted to AI assistants.
AI SOV is one of the six core AI visibility metrics that tell you whether AI recommends your brand — the competitive one. The other five describe you in isolation; share of voice tells you who is eating your lunch.
AI share of voice vs. mention rate: not the same number
These two get conflated constantly, including by tool vendors — and they answer different questions.
| Mention rate | AI share of voice | |
|---|---|---|
| Formula | answers naming you ÷ all answers sampled | your mentions ÷ all tracked brands' mentions |
| Measures | Presence — how often you show up | Competitive position — how much of the pool you own |
The difference is not academic. A brand can hold a 30% mention rate (it appears in 3 of every 10 answers) and only an 18% share of voice, because competitors appear in those same answers more often. Report both, but never swap their labels — executives will anchor on whichever number you show first.
How it differs from traditional share of voice
Traditional SOV divides your advertising spend, impressions or media mentions by the market total. AI SOV keeps the spirit but changes the denominator: the pool is brand mentions inside generated answers, observable only by repeatedly asking the engines and counting what comes back.
Two practical consequences follow. First, there is no exhaustive data feed — every score is a sample, so methodology (prompt set, platforms, sampling window) determines whether your number means anything. Second, the same brand can score 40% on one platform and 12% on another, because each engine retrieves from different sources. Any single blended number hides that spread, which is why serious AI search monitoring reports per-platform SOV alongside the blend.
The AI Share of Voice Formula
The base formula is:
AI SOV (%) = (your brand's mentions ÷ total mentions of all tracked brands) × 100
— counted across the same prompt set, platforms and time window, with each brand counted once per answer. An answer that names you three times still contributes one mention; otherwise verbose answers distort the pool.
The metric comes in three forms, and they diverge sharply on identical data — the worked example below produces 18.75%, 12.5% and 4.4% from the same 2,800 answers. Pick one as your headline metric, label it clearly, and report the others as diagnostics. (Some tools add a sentiment-weighted fourth form; treat it as a diagnostic too — sentiment classification on AI answers is still too noisy to headline.)
Mention-based SOV (the default)
The formula above, applied to brand names appearing anywhere in the answer text. It is the most intuitive and the right default for executive reporting on brand mentions in ChatGPT, Gemini, Perplexity and the rest. Its blind spot: it treats "the market leader is X" and a grudging footnote mention identically.
Position-weighted SOV (rewards being first)
Buyers anchor on the first names in a list, so weighted SOV discounts late mentions. MaxAEO's default weights: 1.0 for a first-position mention, 0.6 for positions 2–3, 0.3 for position 4 or later. (Some teams use harmonic decay — 1/n by position — which is stricter but harder to explain to a CMO.) Apply weights to every brand's mentions, then divide your weighted total by the pool's weighted total.
Citation-based SOV (your content vs. their content)
Instead of brand names, count AI citations — the source links an engine attributes its answer to — and divide your domain's citations by all citations. This measures whether your content feeds the answers, not whether your brand appears in them. The two routinely tell opposite stories, as the worked example below shows.
Worked Example: Calculating AI Share of Voice Across 8 Platforms
Here is a complete calculation, using the structure MaxAEO applies in daily tracking. Scenario: a B2B SaaS brand tracking itself plus four competitors.
- Define the sample. 50 buyer-style prompts × 8 platforms (ChatGPT, Gemini, Perplexity, Claude, Copilot, Grok, Google AI Mode, AI Overviews) = 400 answers per day. Sampled daily for 7 days = 2,800 answers.
- Count answers naming you. Your brand appears in 840 of 2,800 answers → mention rate = 30%.
- Count the full pool. Across all five brands, answers contain 4,480 brand mentions (about 1.6 brands per answer — most answers name several).
- Apply the formula. 840 ÷ 4,480 = 18.75% mention-based AI SOV.
- Weight by position. Your 840 mentions split into 126 first-position, 336 in positions 2–3, 378 in position 4+. Weighted score: (126 × 1.0) + (336 × 0.6) + (378 × 0.3) = 441. The full pool's 4,480 mentions weight to 3,528. Weighted SOV: 441 ÷ 3,528 = 12.5%.
- Check citations. Your domain appears in 84 of 1,920 total cited sources → citation SOV = 4.4%.
| Metric | Calculation | Score |
|---|---|---|
| Mention rate | 840 ÷ 2,800 answers | 30% |
| Mention-based AI SOV | 840 ÷ 4,480 mentions | 18.75% |
| Position-weighted SOV | 441 ÷ 3,528 weighted pool | 12.5% |
| Citation-based SOV | 84 ÷ 1,920 citations | 4.4% |
To rebuild this yourself, log one spreadsheet row per answer with columns for date, platform, prompt, each brand's presence (0/1), best position and cited domains. Every score above falls out of a pivot table — no tooling required.
Read the gaps, not just the scores. Raw SOV (18.75%) near parity says you're in the game. Weighted SOV falling to 12.5% says you're mentioned often but rarely first — your 126 first-position mentions are about 5% of the roughly 2,500 answers that named any brand at all. Citation SOV at 4.4% says the engines describe you using other people's pages. Each gap maps to a different fix, which is exactly why one blended number is never enough.
What Is a Good AI Share of Voice? Benchmarks From 2.4M Answers
A good AI share of voice is anything above parity — 100 divided by the number of brands tracked — and an excellent one is 2× parity or more. In a 5-brand set, parity is 20%: scoring 25% means you punch above your weight; 40%+ means AI treats you as the category default.
Absolute thresholds without that context are meaningless, which is why we normalize. Parity index = your SOV ÷ (100 ÷ N brands). The 18.75% from the worked example is a 0.94 parity index — fractionally below its fair share. The same 18.75% in a 12-brand set would be a 2.25 index: dominant.
Benchmarks by category (MaxAEO tracking data)
The distributions below come from MaxAEO's competitor benchmarking corpus: 412 competitive sets across 38 B2B software categories, 2.4 million answers sampled daily across 8 platforms, January 1 – May 31, 2026. Mentions counted once per answer; sets contain 5–15 brands (median 9).
| Category | Median SOV (all brands) | Top-quartile threshold | Median category leader |
|---|---|---|---|
| CRM & sales tech | 6.8% | ≥ 13.5% | 28.4% |
| Project management | 6.1% | ≥ 14.2% | 31.7% |
| Cybersecurity | 5.2% | ≥ 10.9% | 23.8% |
| Marketing & analytics | 6.5% | ≥ 12.8% | 26.9% |
| HR & payroll | 7.9% | ≥ 15.1% | 33.6% |
| Developer tools | 5.7% | ≥ 11.6% | 27.3% |
The headline finding is concentration: in the median competitive set, the top three brands capture 58% of all mentions (middle 80% of sets: 41–74%). AI answers compress markets harder than search rankings do — there is no page two. The median tracked brand sits around 0.6× parity, because leaders absorb the surplus.
Rough tiers to calibrate against
- Below 0.5× parity — effectively invisible; AI knows the category but not you.
- 0.5–1× parity — contender; you surface, competitors surface more.
- 1–2× parity — shortlist regular; you appear in most relevant answers.
- Above 2× parity — category default; engines volunteer you unprompted.
One caution from the same dataset: scores are set-relative, so adding or removing one competitor mechanically moves everyone's SOV. Freeze your competitor set before you start trending the number, and annotate any changes to it in your reporting.
How to Measure AI Share of Voice Step by Step
You can produce a defensible score manually before buying an AI visibility tool. The method mirrors what platforms automate:
- Freeze a competitor set (5–10 brands). Include who AI actually names, not just who you consider rivals — run a few category prompts first and see every brand AI recommends before yours.
- Build a prompt set of 30–100 queries that mirror real buyer phrasing across the funnel — comparison, "best X for Y," alternatives, pricing. A weak prompt set is the #1 source of garbage scores; here's how to build a prompt set that mirrors what buyers actually ask.
- Sample answers across platforms for at least 7 days. Use clean sessions — no chat history, memory off, consistent locale — because personalization contaminates counts. If you can't cover all eight platforms, start with ChatGPT, Google's AI surfaces (AI Overviews and AI Mode) and Perplexity, and add Copilot if you sell to enterprises.
- Tally mentions once per answer per brand, recording position (1st, 2–3, 4+) and any cited sources while you're there.
- Compute and report three numbers together: mention rate, mention-based SOV, weighted SOV — plus the parity index so stakeholders can interpret the score.
When to move from a spreadsheet to a tracking tool
Manual tracking holds up at one competitive set, three platforms and weekly sampling. It breaks at 8 platforms × daily sampling × multiple brand sets — the point where agencies and in-house teams move LLM brand tracking to an automated platform. Whatever tool you evaluate, require four things:
- Per-platform SOV, not just a blended score
- Position and cited-source capture, not bare mention counts
- Daily sampling rolled into 7- and 28-day views
- Alerts pegged to your baseline volatility, not arbitrary fixed thresholds
Why Your AI Share of Voice Moves Week to Week
A single-day AI SOV reading is noise wearing a suit. The same prompt re-asked tomorrow can return a different shortlist — engines re-retrieve sources, models update, and answers vary across sessions. In MaxAEO's tracking, a brand's single-day SOV deviates from its 30-day average by ±4.2 points (median); a 7-day rolling window cuts that to ±1.3 points.
Practical rules that follow:
- Never report a one-day score. Use 7-day rolling minimum; 28-day for board decks.
- Treat step-changes differently from drift. A sudden 6-point move usually means a model update or a newly-ingested source, not your campaign. Our data on how often AI answers change across 8 platforms shows which engines churn most.
- Set alert thresholds relative to your baseline volatility, not arbitrary round numbers.
How to Raise Your AI Share of Voice
Improving the score means changing what engines retrieve when your category comes up. The levers, in rough order of observed impact:
- Win the pages AI already cites. Engines lean on comparison posts, review sites and "best of" roundups. Across our benchmark corpus, the most common move shared by brands that gained 5+ SOV points in a quarter was new placement on a third-party roundup the engines were already citing — faster impact than anything on your own domain.
- Publish comparison-shaped content on your site — honest alternatives pages, feature tables, pricing clarity. This is the core of answer engine optimization: structure content so an engine can lift a complete, attributable answer from it.
- Standardize your entity description everywhere (site, LinkedIn, directories, Wikipedia-adjacent sources), so models converge on one crisp definition of what you do. Inconsistent descriptions fragment your mentions.
- Fix factual errors at the source. Wrong pricing or a mischaracterized ICP in AI answers suppresses recommendations even when you're mentioned.
- Track weekly and attribute changes. Generative engine optimization without measurement is guesswork; the brands that climb in our benchmarks treat SOV like a conversion rate — hypothesis, change, measured delta.
Sustained execution here is how brands get recommended by ChatGPT and its peers rather than merely mentioned by them.
Frequently Asked Questions
What's the difference between AI share of voice and mention rate?
Mention rate divides answers naming you by total answers sampled — it measures presence. AI share of voice divides your mentions by all tracked brands' mentions — it measures competitive position. A brand can have a high mention rate and a mediocre SOV when competitors appear in the same answers more prominently or more often.
How many prompts do I need for a reliable AI share of voice score?
Use at least 50 prompts sampled over 7+ days. In MaxAEO's tracking data, smaller or single-day samples produce day-to-day swings of ±4 points or more, which is wider than most real month-over-month changes — meaning you'd mostly be reporting noise.
Is AI share of voice the same across ChatGPT, Gemini and Perplexity?
No. Each engine retrieves from different sources, so per-platform scores routinely diverge by 15–25 points for the same brand. Report a blended score for trend lines, but diagnose and fix problems per platform — the sources you need to win differ on each.
Which AI platforms should you track for share of voice?
Weight platforms by where your buyers actually research. For most B2B categories that means ChatGPT, Google AI Overviews/AI Mode and Perplexity as the core, with Microsoft Copilot added for enterprise buying committees. Track at least three so one engine's retrieval quirks don't dominate your trend line.
What is a good AI share of voice for a startup?
Benchmark against parity (100 ÷ brands in your set), not against leaders. New entrants in MaxAEO's benchmark corpus typically start below 0.5× parity — under ~5% in a 10-brand set. Reaching parity within two quarters is a strong trajectory; matching a 30%-SOV incumbent immediately is not a realistic target.
This article was created with AI assistance and reviewed by a human editor.